Digital Image Forensics Based On PRNU And Multi-fractal Spectrum

With the development of digital image acquisition equipment and digital image processing technologies,image retouching is simplified.However,as digital images enrich our daily life and bring convenient to our work,they also lead to digital image security issues.Thus,the research of digital image forensics is of great significance to fight against image forgery and other criminal acts in cyber space.Originated from computer forensics,digital image forensics is the process of collecting,analyzing the forensics data and presenting it to the court.The core of digital forensics is to verify the source of an image,the integrity of it and whe ther it has been tampered based on its features.The classical digital image forensics methods are reviewed in this paper and the theoretical basis of the studies in discriminating natural images and computer generated graphics is discussed.Based on the e xisting methods,the paper focuses on the color filter array(CFA)interpolation and photo response non-uniformity noise(PRNU)characteristics of images and the multi-fractal spectral characteristics of residual images.Then,two novel methods are propose d to identify natural images and computer generated graphics.The main works of the paper are briefly described as follows:1)An identification method based on the features of the impact of CFA interpolation on the local correlation of PRNU is proposed.As CFA interpolation generally exists in the generation of natural images and it imposes influence on the local correlation of PRNU,the differences between the PRNU correlations of natural images and those of computer generated graphics are investigated.Then,the histogram features of PRNU are extracted and a support vector machine(SVM)classifier is used to discriminate natural images and computer generated graphics.Experimental results and analysis show that it can achieve an average identification accuracy of 99.43%,and it is robust against scaling,JPEG compression,rotation and additive noise.2)A method of discriminating natural images and computer generated graphics based on multi-fractal and regression analysis is proposed.Here,residual images of natural images and computer generated graphics are analyzed in detail and the texture difference of them is studied.Moreover,the difference between the fitting degree of the regression model of natural images and computer generated graphics is investigated.Based on these differences,24 dimensions of features are extracted and SVM classifier is used to discriminate natural images and computer generated graphics.Experimental results and analysis show that it can achieve an ave rage identification accuracy of 98.69%,and it is robust against JPEG compression,rotation,additive noise and image resizing.It can be concluded that the proposed methods can effectively discriminate natural images and computer generated graphics.It ha s great potential to be used in image source pipeline forensics.